Researchers led by Google have shown that the optical pulse sensor and motion detector in a consumer smartwatch can flag people likely to have hypertension, with no cuff and no calibration. In a prospective study of 196 adults, the watch detected more confirmed cases than routine office readings, at the same specificity. Published in Nature, the work is early feasibility research: the sample is small and American, and the watch still misses about a third of cases.
High blood pressure is often called a silent killer because most people who have it feel nothing. Over half of the world’s more than one billion adults with hypertension do not know they have it, according to WHO figures cited by the researchers. A study published in Nature on 5 November suggests that a device already worn by hundreds of millions of people could help find them.
The researchers, from Google Research with collaborators including the University of Washington and Beth Israel Deaconess Medical Center in Boston, developed a model called AI-PPG-ACC-HTN. It does not measure blood pressure. Instead, it reads signals that a smartwatch collects routinely and estimates whether the wearer is likely to have hypertension.
How it works
Most smartwatches carry two relevant sensors:
- A photoplethysmography (PPG) sensor: the small green light on the underside, which tracks the pulse by measuring how much light the blood in the wrist absorbs with each heartbeat
- An accelerometer, which measures movement
The shape of the pulse waveform carries information about the arteries. Stiffer arteries, a hallmark of sustained high blood pressure, change how fast the pressure wave travels and how it reflects back from the body’s branching vessels. The authors say the model draws on features such as pulse width, wave reflection and ballistocardiographic signals, the tiny body movements produced by each heartbeat. It also uses the wearer’s height, as a stand-in for the length of the arterial path.
The team first trained the model by self-supervised learning on 492,347 hours of unlabelled sensor data from 2,109 people, letting it learn the structure of the signals without being told who had hypertension. After seven days of ordinary wear, the finished model produces a risk score.
The test
To evaluate it, the researchers recruited 196 adults at five sites in the United States. None had known cardiovascular disease or took blood-pressure medication. Each wore the watch and also underwent 24-hour ambulatory blood-pressure monitoring (ABPM), in which a cuff inflates automatically through the day and night. ABPM is the clinical reference standard because a single reading in a clinic can be misleading. Some people’s pressure rises in the doctor’s office (“white-coat hypertension”), while others have high pressure that is normal in the clinic (“masked hypertension”).
| Comparison | Sensitivity | Specificity |
| Watch, initial screening | 65.8% | 90.0% |
| Office measurement | 55.3% | 90.0% |
| Watch, confirmation step (48 people with high office readings) | 78.4% | 90.9% |
| Repeat office measurement (same 48) | 67.6% | 63.6% |
| Multi-day home cuff monitoring (same 48) | 89.2% | 81.8% |
Sensitivity is the share of people with hypertension whom a test correctly flags; specificity is the share of people without it whom the test correctly clears. At the same 90% specificity, the watch detected more cases than office screening did. Among people with high office readings, it confirmed hypertension more reliably than a repeat office visit, although multi-day home cuff monitoring was more sensitive still.
What the researchers claim, and what they do not
Google’s own description is careful. It says the tool is “not intended to replace standard clinical arm cuffs” and calls the work “early feasibility research”. The company also states that it is “not indicative of any existing or upcoming commercial product feature”, although its blog says the research “helped inform” a blood-pressure trends feature in its health app. NSH notes the tension between those two statements.
The proposed use is opportunistic screening. A watch could flag a wearer whose signals suggest hypertension and prompt them to get a proper cuff measurement. Because the model works in the background over days, it may capture a truer picture than a single clinic reading.
The limitations
- Small sample. There were 196 participants, and only 48 in the confirmation analysis. Larger studies often show weaker performance.
- Narrow population. All sites were in the United States, and people already on treatment or with cardiovascular disease were excluded. Performance in South Asian, older or multimorbid populations is unknown.
- Missed cases. At screening, about one in three people with hypertension were not flagged. A reassuring watch should not stop anyone from having their pressure checked.
- One device. The study used a single, unnamed watch model. Cheaper devices with lower-quality sensors may perform differently.
- No regulatory status. No clearance by any regulator has been reported, and the study was run by the company that developed the model.
What it means for India
The potential relevance to India is large. The ICMR-INDIAB study, published in The Lancet Diabetes & Endocrinology in 2023, estimated that about 315 million Indian adults, about 35.5%, have hypertension. Many are undiagnosed, particularly in rural areas and among younger adults who rarely visit a doctor. India’s national programme for non-communicable diseases screens adults over 30 at health and wellness centres, but coverage is incomplete.
India is also one of the world’s largest markets for smartwatches and fitness bands, many of them low-cost models. If passive screening of this kind were validated in Indian populations and on affordable devices, it could direct people towards cuff testing earlier. Those are significant conditions. Until such validation exists, Indian users should treat any blood-pressure “insight” from a wearable as a prompt, not a diagnosis. A validated arm cuff remains the standard. NSH’s 5 November report on potassium-enriched salt is a reminder that the most effective measures against hypertension are still dietary and clinical.
